Agentic system

Definition

An agentic system combines one or more agents with deterministic software, tools, data, environments, policies, verification, evidence, and human decision rights. The system—not the model alone—is the correct unit for architecture, risk classification, evaluation, and accountability.

The system boundary is increasingly recognized, but product architectures and assurance patterns remain diverse.

Why it matters

The adjective “agentic” describes where control flow is partly model-directed.

A useful design names which choices are dynamic, which effects are mechanically bounded, and which decisions remain human.

System anatomy

Agent layer

Model-directed planning and action.

Deterministic layer

Software boundaries, transactions, limits, and enforcement.

Accountability layer

Named owners, verifiers, and release authorities.

Important distinctions

AI-enabled application

An application may use a model without allowing it to choose a multi-step path or take actions.

AI agent

The agent is one operating component; the agentic system includes every dependency and governing boundary.

Implementation signals

• Draw the system boundary around every model, tool, identity, datastore, verifier, and human authority
• Evaluate end-to-end outcomes and side effects rather than model responses alone

Failure modes

• Treating the model as the whole product
• Leaving vendor tools or human work outside the risk boundary

Sources and further study

1. Anthropic — Trustworthy agents in practice

   Describes a self-directed plan–act–observe–adjust loop and the engineering layers needed to make longer-horizon agents more trustworthy.

   Use in this library: Primary definition. This first-party synthesis describes design directions and failure surfaces; it is not a certification scheme or evidence that every described control is sufficient.

   guidance · guidance · Published 2026-04-09
2. Anthropic — Building effective agents

   A foundational distinction between fixed, code-orchestrated workflows and systems in which a model dynamically directs its process and tool use.

   Use in this library: Primary definition. The workflow-versus-agent distinction is a useful architecture boundary, not a universal taxonomy or proof that autonomous execution is preferable.

   engineering · engineering · Published 2024-12-19
3. NIST — Artificial Intelligence Risk Management Framework 1.0

   A consensus-based foundation for governing, mapping, measuring, and managing AI risk according to context, impact, and organizational priorities.

   Use in this library: Standard or protocol. A specification defines an interface or control pattern; conformance alone does not establish authorization, security, or fitness for a use case.

   standard · standard · Published 2023-01-26